In recent years, Wi-Fi-based human activity sensing has emerged as a prominent research domain, attracting considerable attention across both academic and industrial communities. While the Fresnel zone model has served as a fundamental theoretical framework for Wi-Fi sensing systems, its explanatory power proves inadequate when applied to complex real-world environments. Although numerous studies have attempted to establish theoretical models for cross-link and non-line-of-sight (NLOS) scenarios, a unified theoretical framework for device-free tracking in complex environments remains lacking. To fill this gap, we introduce MetaTrack , which is designed to achieve device-free tracking in complex scenarios in the form of sensing-informed lightweight neural network modules. In this paper, our contribution is twofold: 1) We propose Tracking Heat Zone (THZ), a novel representation that describes wireless signal distributions in diverse scenarios, enabling adaptive modeling of arbitrary environments. 2) We design a standardized sensing-informed modular pipeline that effectively translates THZ representations into lightweight neural network. Such a customizable solution is not only easily deployed, but also flexible to adapt to different kinds of complex scenarios. Experiments in several environment settings show that MetaTrack achieves high-precision tracking not only in ideal scenarios with a median error of 0.48 m , but also in complex environments with a median error 0.52 m , demonstrating its technical strength where it matters most.
Meng et al. (2025) studied this question.